quash-mcp 0.2.13__tar.gz → 0.2.14__tar.gz

{quash_mcp-0.2.13 → quash_mcp-0.2.14}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: quash-mcp
-Version: 0.2.13
+Version: 0.2.14
 Summary: Model Context Protocol server for Quash - AI-powered mobile automation agent
 Project-URL: Homepage, https://quashbugs.com
 Project-URL: Repository, https://github.com/quash/quash-mcp

{quash_mcp-0.2.13 → quash_mcp-0.2.14}/local_test.py

@@ -13,13 +13,15 @@ from quash_mcp.models import SessionDTO, ConfigInfo, UIStateInfo
 async def main():
     # Initialize the session state
     state = get_state()
-    state.config["api_key"] = "mhg_7S01mxJUB7q6k-sbFQeScYTX6ae3pehI"
+    state.config["api_key"] = "mhg_AMncG3jkTtEntd8vTqJk3nHl6rnxceRr"
     state.device_serial = "emulator-5554"
     state.portal_ready = True  # Assume portal is ready for local testing
 
     # Define the task
-    task = "Open chrome and open quashbugs.com website"
-    # task = "Open settings and scroll 3 times."
+    # task = "Open the google search app and search for bakeries."
+    task = "Open settings and scroll 3 times."
+    # task = "Go home"
+    # task = "Open chrome, search for quashbugs.com and scroll down."
 
     # Define a progress callback
     def progress_callback(message):

{quash_mcp-0.2.13 → quash_mcp-0.2.14}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "quash-mcp"
-version = "0.2.13"
+version = "0.2.14"
 description = "Model Context Protocol server for Quash - AI-powered mobile automation agent"
 readme = "README.md"
 license = {text = "MIT"}

{quash_mcp-0.2.13 → quash_mcp-0.2.14}/quash_mcp/backend_client.py

@@ -20,7 +20,7 @@ class BackendClient:
 
     def __init__(self):
         # Get backend URL from environment variable, default to production backend
-        self.base_url = os.getenv("MAHORAGA_BACKEND_URL", "https://mcpbe.quashbugs.com")
+        self.base_url = os.getenv("MAHORAGA_BACKEND_URL", "http://localhost:8000")
         self.timeout = 300.0  # 5 minutes for long-running LLM calls
         logger.info(f"🔧 Backend client initialized: URL={self.base_url}")
 
@@ -35,7 +35,8 @@ class BackendClient:
             Dict with validation result:
             {
                 "valid": bool,
-                "user": {"email": str, "name": str, "credits": float},
+                "user": {"email": str, "name": str},
+                "organization_credits": float,
                 "openrouter_api_key": str,
                 "error": str (if invalid)
             }

{quash_mcp-0.2.13 → quash_mcp-0.2.14}/quash_mcp/tools/execute_v3.py

@@ -17,6 +17,7 @@ from ..state import get_state
 from ..backend_client import get_backend_client
 from ..device.state_capture import get_device_state
 from ..device.adb_tools import AdbTools
+import logging
 
 # Import mahoraga components for tool functions
 try:
@@ -236,12 +237,12 @@ async def execute_v3(
 
     # Check credits
     user_info = validation_result.get("user", {})
-    credits = user_info.get("credits", 0)
+    organization_credits = validation_result.get("organization_credits", 0)
 
-    if credits <= 0:
+    if organization_credits <= 0:
         return {
             "status": "error",
-            "message": f"❌ Insufficient credits. Current balance: ${credits:.2f}",
+            "message": f"❌ Insufficient credits. Current balance: ${organization_credits:.2f}",
             "user": user_info
         }
 
@@ -250,7 +251,7 @@ async def execute_v3(
         if progress_callback:
             progress_callback(message)
 
-    log_progress(f"✅ API Key validated - Credits: ${credits:.2f}")
+    log_progress(f"✅ API Key validated - Credits: ${organization_credits:.2f}")
     log_progress(f"👤 User: {user_info.get('name', 'Unknown')}")
     log_progress(f"🚀 Starting task: {task}")
     log_progress(f"📱 Device: {state.device_serial}")
@@ -315,6 +316,15 @@ async def execute_v3(
                 # Add wrapped function to globals so code can call it directly
                 executor_globals[tool_name] = make_printing_wrapper(tool_function)
 
+            # Override the 'complete' function to be a no-op
+            # The backend already handles completion via the 'completed' flag
+            def complete_no_op(success=True, reason=""):
+                """No-op wrapper for complete() - completion is handled by backend."""
+                print(f"complete() called: success={success}, reason='{reason}'")
+                return None
+
+            executor_globals['complete'] = complete_no_op
+
             log_progress(f"🔧 Loaded {len(filtered_tools)} tool functions: {list(filtered_tools.keys())}")
         except Exception as e:
             log_progress(f"⚠️ Warning: Could not load tool functions: {e}")
@@ -402,50 +412,8 @@ async def execute_v3(
                 log_progress(f"🤔 Reasoning: {reasoning}")
 
 
-            # 3. Check if task is complete BEFORE executing action
-            if step_result.get("completed", False):
-                success = step_result.get("success", False)
-                final_message = step_result.get("final_message", "Task completed")
-
-                duration = time.time() - start_time
-
-                if success:
-                    log_progress(f"✅ Task completed successfully!")
-                else:
-                    log_progress(f"❌ Task marked as failed")
-
-                # Finalize session on backend
-                finalize_result = await backend.finalize_session(session=session)
-
-                if success:
-                    log_progress(f"✅ Task completed successfully in {len(session.steps)} steps")
-                    log_progress(f"💰 Usage: {finalize_result.get('total_tokens', {}).get('total')} tokens, ${finalize_result.get('total_cost', 0):.4f}")
-
-                    return {
-                        "status": "success",
-                        "steps_taken": len(session.steps),
-                        "final_message": final_message,
-                        "message": f"✅ Success: {final_message}",
-                        "tokens": finalize_result.get("total_tokens"),
-                        "cost": finalize_result.get("total_cost"),
-                        "duration_seconds": duration
-                    }
-                else:
-                    log_progress(f"❌ Task failed: {final_message}")
-                    log_progress(f"💰 Usage: {finalize_result.get('total_tokens', {}).get('total')} tokens, ${finalize_result.get('total_cost', 0):.4f}")
-
-                    return {
-                        "status": "failed",
-                        "steps_taken": len(session.steps),
-                        "final_message": final_message,
-                        "message": f"❌ Failed: {final_message}",
-                        "tokens": finalize_result.get("total_tokens"),
-                        "cost": finalize_result.get("total_cost"),
-                        "duration_seconds": duration
-                    }
-
-
-            # 4. Execute action locally (only if task is not complete)
+            # 3. Execute action locally FIRST (if provided)
+            # NOTE: Backend should have already removed complete() from the code
             if code and action_type == "execute_code":
                 log_progress(f"⚡ Executing action...")
 
@@ -514,6 +482,8 @@ async def execute_v3(
 
                     session.chat_history.append(ChatHistoryMessage(role="user", content=f"Execution Result:\n```\n{feedback}\n```"))
 
+                    # Introduce a small delay to allow UI effects to settle before checking completion
+                    time.sleep(1.0) # Added delay
 
                 except Exception as e:
                     error_msg = f"Error during execution: {str(e)}"
@@ -526,6 +496,48 @@ async def execute_v3(
                 session.chat_history.append(ChatHistoryMessage(role="user", content="No code was provided. Please provide code to execute."))
 
 
+            # 4. Check if task is complete AFTER executing action
+            if step_result.get("completed", False):
+                success = step_result.get("success", False)
+                final_message = step_result.get("final_message", "Task completed")
+
+                duration = time.time() - start_time
+
+                if success:
+                    log_progress(f"✅ Task completed successfully!")
+                else:
+                    log_progress(f"❌ Task marked as failed")
+
+                # Finalize session on backend
+                finalize_result = await backend.finalize_session(session=session)
+
+                if success:
+                    log_progress(f"✅ Task completed successfully in {len(session.steps)} steps")
+                    log_progress(f"💰 Usage: {finalize_result.get('total_tokens', {}).get('total')} tokens, ${finalize_result.get('total_cost', 0):.4f}")
+
+                    return {
+                        "status": "success",
+                        "steps_taken": len(session.steps),
+                        "final_message": final_message,
+                        "message": f"✅ Success: {final_message}",
+                        "tokens": finalize_result.get("total_tokens"),
+                        "cost": finalize_result.get("total_cost"),
+                        "duration_seconds": duration
+                    }
+                else:
+                    log_progress(f"❌ Task failed: {final_message}")
+                    log_progress(f"💰 Usage: {finalize_result.get('total_tokens', {}).get('total')} tokens, ${finalize_result.get('total_cost', 0):.4f}")
+
+                    return {
+                        "status": "failed",
+                        "steps_taken": len(session.steps),
+                        "final_message": final_message,
+                        "message": f"❌ Failed: {final_message}",
+                        "tokens": finalize_result.get("total_tokens"),
+                        "cost": finalize_result.get("total_cost"),
+                        "duration_seconds": duration
+                    }
+
 
         # Max steps reached
         log_progress(f"⚠️ Reached maximum steps ({max_steps})")